Accelerate Attention with FlashAttention-3: New Capabilities and Performance
FlashAttention-3 significantly accelerates attention in AI models, achieving 1.2 PFLOPS with FP8 and improving GPU performance.
#Performance (3)
Accelerate inference with torch.compile caching
Torch.compile caching accelerates model boot times in PyTorch by 2-3 times.
Introducing Gemini 3.1 Flash-Lite for Scalable Intelligence
Google has introduced Gemini 3.1 Flash-Lite, a fast and economical model for developers and enterprises.
·
30 просмотров